Ultra-Efficient Si3N4 MZIs With BaTiO3 as Weight Elements for Neuromorphic Photonics

This paper introduces a highly energy efficient photonic Mach-Zehnder interferometer (MZI) on the silicon nitride (Si₃N₄) platform, featuring hybrid Si₃N₄-BaTiO₃ (BTO) electro-optic phase shifters (PS) for photonic neural networks (PNNs). The experimental results of the fabricated devices reveal low power consumption of only 121 nW for a π phase shift, extinction ratio surpassing 27 dB, and exceptionally low $\text{V}_{{\rm{\pi }}}\times \text{L}$ of 2.52 Vmm for a 2.1 mm long PS with insertion losses of 4.65 dB. This value can go down to 0.95 dB with proper processing of the BTO layer for fine etching according to 3D-FDTD simulations. Furthermore, the MZIs demonstrate rapid rise/fall times below 20 ns for 5 MHz RF square pulse modulation. The low loss and power consumption renders the proposed technology highly suitable for reconfigurable weighting elements in large-scale multi-layer PNNs.